3 ms·
It's nice to see someone tell you that their benchmark is flawed and why. Most try to pass off their benchmark as the end-all-be-all measurement of whatever th
by phsr 16y ago
It's nice to see someone tell you that their benchmark is flawed and why. Most try to pass off their benchmark as the end-all-be-all measurement of whatever they are testing
- JoeAltmaier 16y agoGraphs were also flawed - scale grossly exaggerated to make the point. Why not zero-based vertical scale? See "How to Lie with Statistics".
- masklinn 16y agoIf the scales were flawed, it's in defense of -fomit-frame-pointer: according to his bench, it reduces the cycle count by ~0.3% in 32b, but blows it up by nearly 30% (29.7) in 64b. I have put both graphs on the same image and scaled them correctly, then extended the canvas to the whole scale and finally shrunk the graph back to the original 600px high, this is the result: http://imgur.com/yG3R9 http://imgur.com/yG3R9 64b on the left, 32 on the right, blue is without -fomit and red is with it. On the 32b graph, you can barely discriminate between with and without, whereas on the 64b graph you can very clearly see it. If he lied, his fault is to have dismissed his own findings as less important than they are.
- ice799 16y agogrossly exaggerated? bro, i gave the raw numbers in a table above the graph. if i were trying to lie i wouldn't have given the numbers. i took the graphs as-is from goog docs and put em in my blog post. i just zero-based the vertical scale. you can close your book about lying with stats or whatever the shit.